Publications
Integrative sensor networks, informatics, and modeling for precision and preventative medicine
July 1, 2020
Journal Article
Published in:
IEEE J. Biomed. Health, Vol. 24, No. 7, July 2020, pp. 1858-1859.
Topic:
R&D area:
R&D group:
Summary
The topics of integrative sensor networks, informatics and modeling bring together the tightly coupled and rapidly developing fields of biomedical and health informatics and body sensor networks. Biomedical and health informatics encompasses methods to extract and communicate information from data in order to impact health, healthcare, life sciences and biomedicine. Body sensor networks provide one means to measure the needed data, through continuous monitoring in both clinical and free-living environments. Developments in these areas were highlighted at two co-located conferences: the 2019 IEEE-EMBS International Conferences on Biomedical and Health Informatics (BHI'19) and Wearable and Implantable Body Sensor Networks (BSN'19). BHI and BSN are long-standing IEEE EMBS conferences that provide a forum for researchers and leaders from academia, government and industry to share technical advances and new initiatives in these important areas. Through an open call for this special issue, eleven papers have been included for publication. The majority were presented in an initial form at the 2018 or 2019 BHI and BSN conferences. Nine of the papers were selected through a rigorous peer review. In addition, two keynote speakers from BHI'19 and BSN'19 have provided short position papers.
Summary
The topics of integrative sensor networks, informatics and modeling bring together the tightly coupled and rapidly developing fields of biomedical and health informatics and body sensor networks. Biomedical and health informatics encompasses methods to extract and communicate information from data in order to impact health, healthcare, life sciences and biomedicine...
READ MORE
Deep implicit coordination graphs for multi-agent reinforcement learning [e-print]
June 19, 2020
Journal Article
Published in:
https://arxiv.org/abs/2006.11438
Topic:
R&D area:
R&D group:
Summary
Multi-agent reinforcement learning (MARL) requires coordination to efficiently solve certain tasks. Fully centralized control is often infeasible in such domains due to the size of joint action spaces. Coordination graph based formalization allows reasoning about the joint action based on the structure of interactions. However, they often require domain expertise in their design. This paper introduces the deep implicit coordination graph (DICG) architecture for such scenarios. DICG consists of a module for inferring the dynamic coordination graph structure which is then used by a graph neural network based module to learn to implicitly reason about the joint actions or values. DICG allows learning the tradeoff between full centralization and decentralization via standard actor-critic methods to significantly improve coordination for domains with large number of agents. We apply DICG to both centralized-training-centralized-execution and centralized-training-decentralized-execution regimes. We demonstrate that DICG solves the relative overgeneralization pathology in predatory-prey tasks as well as outperforms various MARL baselines on the challenging StarCraft II Multi-agent Challenge (SMAC) and traffic junction environments.
Summary
Multi-agent reinforcement learning (MARL) requires coordination to efficiently solve certain tasks. Fully centralized control is often infeasible in such domains due to the size of joint action spaces. Coordination graph based formalization allows reasoning about the joint action based on the structure of interactions. However, they often require domain expertise...
READ MORE
Control systems need software security too: cyber-physical systems and safety-critical application domains must adopt widespread effective software defenses
June 1, 2020
Journal Article
Published in:
SIGNAL Mag., 1 June 2020.
Topic:
R&D group:
Summary
Low-level embedded control systems are increasingly being targeted by adversaries, and there is a strong need for stronger software defenses for such systems. The cyber-physical nature of such systems impose real-time performance constraints not seen in enterprise computing systems, and such constraints fundamentally alter how software defenses should be designed and applied. MIT Lincoln Laboratory scientists demonstrated that current randomization-based defenses, which have low average-case overhead, can incur significant worst-case overhead that may be untenable in real-time applications, while some low-overhead enforcement-based defenses have low worst-case performance overheads making them more amenable to real-time applications. Such defenses should be incorporated into a comprehensive resilient architecture with a strategy for failover and timely recovery in the case of a cyber threat.
Summary
Low-level embedded control systems are increasingly being targeted by adversaries, and there is a strong need for stronger software defenses for such systems. The cyber-physical nature of such systems impose real-time performance constraints not seen in enterprise computing systems, and such constraints fundamentally alter how software defenses should be designed...
READ MORE
COVID-19: famotidine, histamine, mast cells, and mechanisms [eprint]
May 23, 2020
Journal Article
Published in:
Research Square, preprint, 23 May 2020.
Topic:
R&D area:
R&D group:
Summary
SARS-CoV-2 infection is required for COVID-19, but many signs and symptoms of COVID-19 differ from common acute viral diseases. Currently, there are no pre- or post-exposure prophylactic COVID-19 medical countermeasures. Clinical data suggest that famotidine may mitigate COVID-19 disease, but both mechanism of action and rationale for dose selection remain obscure. We explore several plausible avenues of activity including antiviral and host-mediated actions. We propose that the principal famotidine mechanism of action for COVID-19 involves on-target histamine receptor H2 activity, and that development of clinical COVID-19 involves dysfunctional mast cell activation and histamine release.
Summary
SARS-CoV-2 infection is required for COVID-19, but many signs and symptoms of COVID-19 differ from common acute viral diseases. Currently, there are no pre- or post-exposure prophylactic COVID-19 medical countermeasures. Clinical data suggest that famotidine may mitigate COVID-19 disease, but both mechanism of action and rationale for dose selection remain...
READ MORE
75,000,000,000 streaming inserts/second using hierarchical hypersparse GraphBLAS matrices
May 18, 2020
Conference Paper
Published in:
2020 IEEE Intl. Parallel and Distributed Processing Symp. Workshops, IPDPSW, 18-22 May 2020.
Topic:
R&D area:
R&D group:
Summary
The SuiteSparse GraphBLAS C-library implements high performance hypersparse matrices with bindings to a variety of languages (Python, Julia, and Matlab/Octave). GraphBLAS provides a lightweight in-memory database implementation of hypersparse matrices that are ideal for analyzing many types of network data, while providing rigorous mathematical guarantees, such as linearity. Streaming updates of hypersparse matrices put enormous pressure on the memory hierarchy. This work benchmarks an implementation of hierarchical hypersparse matrices that reduces memory pressure and dramatically increases the update rate into a hypersparse matrices. The parameters of hierarchical hypersparse matrices rely on controlling the number of entries in each level in the hierarchy before an update is cascaded. The parameters are easily tunable to achieve optimal performance for a variety of applications. Hierarchical hypersparse matrices achieve over 1,000,000 updates per second in a single instance. Scaling to 31,000 instances of hierarchical hypersparse matrices arrays on 1,100 server nodes on the MIT SuperCloud achieved a sustained update rate of 75,000,000,000 updates per second. This capability allows the MIT SuperCloud to analyze extremely large streaming network data sets.
Summary
The SuiteSparse GraphBLAS C-library implements high performance hypersparse matrices with bindings to a variety of languages (Python, Julia, and Matlab/Octave). GraphBLAS provides a lightweight in-memory database implementation of hypersparse matrices that are ideal for analyzing many types of network data, while providing rigorous mathematical guarantees, such as linearity. Streaming updates...
READ MORE
The 2017 Buffalo Area Icing and Radar Study (BAIRS II)
May 14, 2020
Project Report
Published in:
MIT Lincoln Laboratory Report ATC-447
Topic:
R&D area:
R&D group:
Summary
The second Buffalo Area Icing and Radar Study (BAIRS II) was conducted during the winter of 2017. The BAIRS II partnership between Massachusetts Institute of Technology (MIT) Lincoln Laboratory (LL), the National Research Council of Canada (NRC), and Environment and Climate Change Canada (ECCC) was sponsored by the Federal Aviation Administration (FAA). It is a follow-up to the similarly sponsored partnership of the original BAIRS conducted in the winter of 2013. The original BAIRS provided in situ verification and validation of icing and hydrometeors, respectively, within the radar domain in support of a hydrometeor-classification-based automated icing hazard algorithm. The BAIRS II motivation was to: --Collect additional in situ verification and validation data, --Probe further dual polarimetric radar features associated with icing hazard, --Provide foundations for additions to the icing hazard algorithm beyond hydrometeor classifications, and --Further characterize observable microphysical conditions in terms of S-band dual polarimetric radar data. With BAIRS II, the dual polarimetric capability is provided by multiple Next Generation Weather Radar (NEXRAD) S-band radars in New York State, and the verification of the icing hazard with microphysical and hydrometeor characterizations is provided by NRC's Convair-580 instrumented research plane during five icing missions covering about 21 mission hours. The ability to reliably interpret the NEXRAD dual polarization radar-sensed thermodynamic phase of the hydrometeors (solid, liquid, mix) in the context of cloud microphysics and precipitation physics makes it possible to assess the icing hazard potential to aviation. The challenges faced are the undetectable nature of supercooled cloud droplets (for Sband) and the isotropic nature of Supercooled Large Drops (SLD). The BAIRS II mission strategy pursued was to study and probe radar-identifiable, strongly anisotropic crystal targets (dendrites and needles) with which supercooled water (and water saturated conditions) are physically linked as a means for dual polarimetric detection of icing hazard. BAIRS II employed superior optical array probes along with state and microphysical instrumentation; and, using again NEXRAD-feature-guided flight paths, was able to make advances from the original BAIRS helpful to the icing algorithm development. The key findings that are given thorough treatment in this report are: --Identification of the radar-detectable "crystal sandwich" structure from two anisotropic crystal types stratified by in situ air temperature in association with varying levels of supercooled water --with layer thicknesses observed to 2 km, --over hundred-kilometer scales matched with the mesoscale surveillance of the NEXRAD radars, --Development and application of a multi-sensor cloud phase algorithm to distinguish between liquid phase, mixed phase, and glaciated (no icing) conditions for purposes of a "truth" database and improved analysis in BAIRS II, --Development of concatenated hydrometeor size distributions to examine the in situ growth of both liquid and solid hydrometeors over a broad size spectrum; used, in part, to demonstrate differences between maritime and continental conditions, and --The Icing Hazard Levels (IHL) algorithm’s verification in icing conditions is consistent with previous work and, new, is documented to perform well when indicating "glaciated" (no icing) conditions.
Summary
The second Buffalo Area Icing and Radar Study (BAIRS II) was conducted during the winter of 2017. The BAIRS II partnership between Massachusetts Institute of Technology (MIT) Lincoln Laboratory (LL), the National Research Council of Canada (NRC), and Environment and Climate Change Canada (ECCC) was sponsored by the Federal Aviation...
READ MORE
Kawasaki disease, multisystem inflammatory syndrome in children: antibody-induced mast cell activation hypothesis
May 12, 2020
Journal Article
Published in:
J Pediatrics & Pediatr Med. 2020; 4(2): 1-7
Topic:
R&D area:
R&D group:
Summary
Multisystem Inflammatory Syndrome in Children (MIS-C) is appearing in infants, children, and young adults in association with COVID-19 (coronavirus disease 2019) infections of SARS-CoV-2. Kawasaki Disease (KD) is one of the most common vasculitides of childhood. KD presents with similar symptoms to MIS-C especially in severe forms such as Kawasaki Disease Shock Syndrome (KDSS). The observed symptoms for MIS-C and KD are consistent with Mast Cell Activation Syndrome (MCAS) characterized by inflammatory molecules released from activated mast cells. Based on the associations of KD with multiple viral and bacterial pathogens, we put forward the hypothesis that KD and MIS-C result from antibody activation of mast cells by Fc receptor-bound pathogen antibodies causing a hyperinflammatory response upon second pathogen exposure. Within this hypothesis, MIS-C may be atypical KD or a KD-like disease associated with SARS-CoV-2. We extend the mast cell hypothesis that increased histamine levels are inducing contraction of effector cells with impeded blood flow through cardiac capillaries. In some patients, pressure from impeded blood flow, within cardiac capillaries, may result in increased coronary artery blood pressure leading to aneurysms, a well-known complication in KD.
Summary
Multisystem Inflammatory Syndrome in Children (MIS-C) is appearing in infants, children, and young adults in association with COVID-19 (coronavirus disease 2019) infections of SARS-CoV-2. Kawasaki Disease (KD) is one of the most common vasculitides of childhood. KD presents with similar symptoms to MIS-C especially in severe forms such as Kawasaki...
READ MORE
Bayesian estimation of PLDA with noisy training labels, with applications to speaker verification
May 4, 2020
Conference Paper
Published in:
2020 IEEE Intl. Conf. on Acoustics, Speech and Signal Processing, ICASSP, 4-8 May 2020.
Topic:
R&D area:
Summary
This paper proposes a method for Bayesian estimation of probabilistic linear discriminant analysis (PLDA) when training labels are noisy. Label errors can be expected during e.g. large or distributed data collections, or for crowd-sourced data labeling. By interpreting true labels as latent random variables, the observed labels are modeled as outputs of a discrete memoryless channel, and the maximum a posteriori (MAP) estimate of the PLDA model is derived via Variational Bayes. The proposed framework can be used for PLDA estimation, PLDA domain adaptation, or to infer the reliability of a PLDA training list. Although presented as a general method, the paper discusses specific applications for speaker verification. When applied to the Speakers in the Wild (SITW) Task, the proposed method achieves graceful performance degradation when label errors are introduced into the training or domain adaptation lists. When applied to the NIST 2018 Speaker Recognition Evaluation (SRE18) Task, which includes adaptation data with noisy speaker labels, the proposed technique provides performance improvements relative to unsupervised domain adaptation.
Summary
This paper proposes a method for Bayesian estimation of probabilistic linear discriminant analysis (PLDA) when training labels are noisy. Label errors can be expected during e.g. large or distributed data collections, or for crowd-sourced data labeling. By interpreting true labels as latent random variables, the observed labels are modeled as...
READ MORE
One giant leap for computer security
April 23, 2020
Journal Article
Published in:
IEEE Security & Privacy, Vol. 18, No. 4, July-August 2020.
Topic:
R&D group:
Summary
Today's computer systems trace their roots to an era of trusted users and highly constrained hardware; thus, their designs fundamentally emphasize performance and discount security. This article presents a vision for how small steps using existing technologies can be combined into one giant leap for computer security.
Summary
Today's computer systems trace their roots to an era of trusted users and highly constrained hardware; thus, their designs fundamentally emphasize performance and discount security. This article presents a vision for how small steps using existing technologies can be combined into one giant leap for computer security.
READ MORE
Discriminative PLDA for speaker verification with X-vectors
April 9, 2020
Journal Article
Published in:
IEEE Signal Processing Letters [submitted]
Topic:
R&D area:
Summary
This paper proposes a novel approach to discrimina-tive training of probabilistic linear discriminant analysis (PLDA) for speaker verification with x-vectors. Model over-fitting is a well-known issue with discriminative PLDA (D-PLDA) forspeaker verification. As opposed to prior approaches which address this by limiting the number of trainable parameters, the proposed method parameterizes the discriminative PLDA (D-PLDA) model in a manner which allows for intuitive regularization, permitting the entire model to be optimized. Specifically, the within-class and across-class covariance matrices which comprise the PLDA model are expressed as products of orthonormal and diagonal matrices, and the structure of these matrices is enforced during model training. The proposed approach provides consistent performance improvements relative to previous D-PLDA methods when applied to a variety of speaker recognition evaluations, including the Speakers in the Wild Core-Core, SRE16, SRE18 CMN2, SRE19 CMN2, and VoxCeleb1 Tasks. Additionally, when implemented in Tensorflow using a modernGPU, D-PLDA optimization is highly efficient, requiring less than 20 minutes.
Summary
This paper proposes a novel approach to discrimina-tive training of probabilistic linear discriminant analysis (PLDA) for speaker verification with x-vectors. Model over-fitting is a well-known issue with discriminative PLDA (D-PLDA) forspeaker verification. As opposed to prior approaches which address this by limiting the number of trainable parameters, the proposed method...
READ MORE